---
title: "FAQ"
description: "Practical answers about discovery, supported execution, accounts and data."
canonical: "https://agenttwin.si/docs/faq/"
---

# FAQ

Practical answers about discovery, supported execution, accounts and data.


## Can I use AgentTwin now?

Yes, for explicitly configured bundled Node workflows using supported operations. Run `npx agenttwin init` with Node.js 22.12+, review discovery, prepare your real files and execute locally against supported HTTP twins, or explicitly choose a cloud sandbox. The dashboard supports individual AgentTwin accounts; users can also configure their own E2B account for direct CLI execution.

## Does it clone my production accounts or every tool?

No. Discovery finds dependency and code signals. Supported services use synthetic state and fresh test credentials. A catalog entry does not imply a working vendor adapter. Provider-native account provisioning and arbitrary production-environment cloning are not implemented.

## Is it only for refunds?

No. The product addresses workflows and agent side effects across systems. Four fault classes ship today: timeout-after-write, blackhole (a dependency going dark mid-call), server-error (repeated failures from a dependency) and duplicate-write (first-class duplicate deliveries, suppressed by idempotency keys). Seeded reproduction and named property checks apply across all of them. Refund scenarios are historical examples.

## What happens to credentials and data?

The operator’s E2B key stays on the API server. Dashboard account tokens remain in tab memory. Sync sends reviewed metadata, not source bodies or credential values. Execution uploads explicitly reviewed application files and fixtures; those files can themselves contain sensitive values, so review the bundle. No production OAuth session is reused. The terminal CLI keeps its saved login and runtime-template config only under `~/.agenttwin/` and does not scan other dotfiles or shell environments for credentials.

## Are execution sandboxes zero-retention?

Optional E2B cloud sandboxes are created fresh for each run. Termination is attempted afterward; cleanup can fail and configured expiration is the backstop. This does not establish a zero-retention guarantee. Submitted source is not kept for recovery. What *is* retained is the evidence: receipts, with recorded events and before/after state, persist in your account (capped at 1,000 receipts per account; entries older than 30 days are pruned on gateway startup and new run submission, not by a scheduled deletion deadline) — that retention is the product, not an accident. There is no self-service delete-on-request pipeline for stored receipts yet; account revocation blocks new runs but does not erase history today.

## Does init automatically run and repair every workflow?

No. Init offers agent-assisted analysis and exports workflow candidates. Inputs, invocation, dependency mappings and assertions need review. Fix & retest exports recorded evidence for your coding agent; a rebuilt bundle and fresh execution are required to verify a repair.

## What is saved in the browser?

Manual declarations, map layout and preferences persist in browser storage. Imported discovery and receipts stay in memory. Account-owned discovery and jobs persist on the API. Tokens are cleared on disconnect or reload.

## Is the prototype a real execution?

The separate prototype is scripted and uses fictional scenarios. The dashboard and CLI operate on actual imported/synced discovery and hosted run receipts. Engineering examples are not customer references.